AI-based digital twin of a building
A building's digital twin is more than just a 3D model. It's a living system, synchronized with the physical structure via sensors, enriched with machine learning predictions, and capable of optimizing engineering systems in real time. Overall operating cost savings are 15-30% compared to pre-implementation levels.
Digital Twin Architecture
Model levels:
- Geometric: BIM model (Revit, IFC) - geometry, building structures, utility networks
- Semantic: building ontology - rooms, zones, equipment, their relationships
- Real-time data: sensors, meters, SCADA - current status
- Predictive: ML models - future state, risks, recommendations
Technology stack:
BIM (Revit/OpenBIM) → IFC конвертация
↓
Knowledge Graph (Apache Jena, Stardog) — граф онтологии здания
↓
IoT Data Platform (ThingsBoard, AWS IoT Core) — сенсорные данные
↓
Digital Twin Platform (Bentley iTwin, Azure Digital Twins, Siemens Xcelerator)
↓
ML/Analytics Engine (Python, PyTorch, scikit-learn)
↓
Dashboard (Grafana, custom React/3D WebGL)
Data sources
Building engineering systems:
- BMS/BAS (Building Management System): HVAC, lighting, elevators
- SCADA: boiler room, chillers, pumping stations
- Protocols: BACnet, KNX, Modbus, LonWorks → MQTT conversion
Sensor:
- Temperature/humidity in the rooms (each zone)
- CO₂ - indicator of human presence and ventilation
- Electricity, heat, water meters (smart meters, ASCUE)
- Occupancy sensors (PIR, video analytics without face recording)
- Access control systems: real-time presence data by zone
External data:
- Weather forecast (Open-Meteo, Yandex.Weather API): input parameter for HVAC
- Calendar: working days, holidays, events in the building
- Electricity tariffs: time-of-use tariffs
Thermal model and HVAC optimization
RC Thermal Network (physical model):
class ThermalZoneModel:
"""
R-C сеть: каждая зона = тепловая ёмкость C
Теплообмен через стены (R_wall), окна (R_window)
"""
def __init__(self, C_zone, R_wall, R_window, R_hvac):
self.C = C_zone # Дж/К
self.R_wall = R_wall # К/Вт
self.R_window = R_window
self.R_hvac = R_hvac
def next_temperature(self, T_zone, T_outdoor, T_supply_air, Q_occupants, dt):
Q_wall = (T_outdoor - T_zone) / self.R_wall
Q_window = (T_outdoor - T_zone) / self.R_window
Q_hvac = (T_supply_air - T_zone) / self.R_hvac
Q_total = Q_wall + Q_window + Q_hvac + Q_occupants
dT = Q_total / self.C * dt
return T_zone + dT
Model calibration: The RC parameters (C, R) are estimated from historical data using Bayesian optimization or scipy.optimize. A Kalman Filter is used for real-time state updates.
MPC (Model Predictive Control):
from scipy.optimize import minimize
def mpc_hvac_optimization(zone_models, weather_forecast_48h, occupancy_forecast,
tariff_schedule, comfort_bounds):
"""
Горизонт: 24-48 часов
Оптимизируемые переменные: setpoints для каждой зоны × каждый час
Цель: минимизация стоимости энергии при соблюдении комфорта
"""
def objective(u):
cost = 0
temperatures = simulate_building(zone_models, u, weather_forecast_48h, occupancy_forecast)
for t, (temps, tariff) in enumerate(zip(temperatures, tariff_schedule)):
energy = compute_energy(u[t])
cost += energy * tariff
return cost
def comfort_constraint(u):
temps = simulate_building(zone_models, u, weather_forecast_48h, occupancy_forecast)
violations = [max(0, comfort_bounds['min'] - t) + max(0, t - comfort_bounds['max'])
for period_temps in temps for t in period_temps]
return -sum(violations)
result = minimize(objective, x0=baseline_setpoints,
constraints={'type': 'ineq', 'fun': comfort_constraint},
method='SLSQP')
return result.x
Savings: MPC reduces HVAC energy consumption by 15-25% vs. PID controllers with fixed setpoints.
Lighting control
Occupancy-driven lighting:
# Прогноз занятости помещений на следующий час
# Ввод: данные СКУД, PIR, CO₂, исторические паттерны по дню недели
occupancy_model = LightGBMClassifier()
predicted_occupancy = occupancy_model.predict_proba(hour_features)
# Диммирование: яркость = f(прогнозируемая занятость + daylight harvesting)
daylight_factor = lux_sensor_outdoor / lux_sensor_indoor
target_illuminance = 500 # люкс для рабочего места
artificial_contribution = max(0, target_illuminance - daylight_factor * outdoor_lux)
dimming_level = artificial_contribution / max_illuminance
DALI control: digital lighting control protocol - individual commands for each luminaire.
Technical condition monitoring
Predictive Maintenance of Engineering Systems:
- Chillers: COP (coefficient of performance) below normal → refrigerant degradation or condenser clogging
- AHU: pressure drop across filters → contamination, replacement according to condition, not according to schedule
- Pumps: vibration + power consumption → bearing wear
- Elevators: time-to-destination, door cycles, motor current - prediction of replacement of cables, door mechanisms
Defect escalation:
defect_severity = {
'low': 'log_in_cmms', # планируемое ТО
'medium': 'schedule_next_maintenance',
'high': 'notify_engineer',
'critical': 'immediate_alert + auto_shutdown_if_safe'
}
Carbon Footprint Analysis
Real-time Carbon Tracking:
- Actual consumption x network carbon intensity (g CO₂/kWh hourly) = current carbon footprint
- Optimization: shifting the flexible load to hours with low carbon intensity (more renewable energy in the network)
- SCOPE 1-2 Reporting for ESG Disclosure
Net Zero Dashboard: Progress towards decarbonization targets: base year → current → projected.
Integration and scaling
Multi-building Portfolio: One Digital Twin Platform → network buildings: shopping centers, office parks, hotel chains. Benchmarking between facilities: which building consumes more energy than the standard under similar conditions.
API integration:
- CMMS (Maximo, 1C:TO): automatic creation of maintenance tasks
- ERP: planned maintenance costs according to forecast
- Tenant Billing: distribution of energy costs among tenants based on actual consumption
Deadlines: BIM → IFC import, basic HVAC data, monitoring dashboard – 6-8 weeks. Thermal RC model, MPC HVAC, occupancy-driven lighting, predictive maintenance – 4-5 months. Full-fledged Digital Twin with carbon tracking, multi-building, CMMS integration – 7-9 months.